Even if the statistical analysis was sound the conclusions don't follow. The "effect" of personalized responses may be due to multi-colinearity: personal letter

Even if the statistical analysis was sound the conclusions don't follow. The "effect" of personalized responses may be due to multi-colinearity: personal letters may be written only in cases when there is a genuine personal connection, so adding personalization may have no effect in other circumstances. There are other issues as well, but that's the most obvious. The whole thing is a cautionary tale of not doing data analysis when you don't have appropriate training and experience.